AI MarketingVideo Blog4 min read
Episode 131 of Brains Bots and Business, “Stop Prompting and Start Planning: The Blueprint for AI Success”, is above. Here’s the written version.
If you tell your AI to “just go do it” without approving a blueprint first, you can burn through your budget and get terrible results. Sandeep’s fix is plan mode: the agent outlines the steps it intends to take, the files it will touch and where its checks are, and you approve that plan before it acts.
It’s the same idea behind Stop Writing Prompts (Do This Instead), taken one step further: stop polishing the prompt, and start reviewing the plan.
Key takeaways
- With a chatbot, a bad answer costs you a click on “regenerate”. With an autonomous agent, a bad run can cost you budget and files.
- A plan exposes the agent’s assumptions, the files it will modify, the tools it will use and where the verification checks are.
- Approving the plan is the guardrail.
What is the “prompting and praying” problem?
For years we’ve been trained to write a prompt and hit enter. In the old chatbot era, if the AI got it wrong, no big deal. You hit regenerate.
Sandeep’s point is that we’re now entering the era of autonomous desktop agents. As he describes them, these AI workers can read your files, control your browser and run unattended for hours. (It’s the shift we covered in The Death of the Chatbot.)
Give an agent like that a vague prompt, then walk away, and it’s like telling a general contractor to “build me a house” without a blueprint. You end up with a very expensive mess. That’s prompting and praying. It’s also the risk behind The Danger of “YOLO Mode”.
What is plan mode for an AI agent?
Sandeep’s view: the smartest sales and marketing leaders are moving from prompting to planning.
In some agent tools, a plan isn’t a vague idea of what might happen. It’s a detailed, step-by-step outline of exactly what the AI intends to do before it takes a single action. Plan mode forces the AI to write that outline first.
A good plan exposes the agent’s assumptions. It tells you which files it will modify, which tools it will use and where the verification checks are. Then you decide whether to let it start.
What does planning look like in practice?
Here’s Sandeep’s example. Say you want your desktop agent to audit and update your entire marketing library for a new product launch.
Force the agent into plan mode first and you get to review the exact route it will take. You place the guardrails. You approve the blueprint. That one step helps you catch runaway processes that burn through your API budget, and the wrong files getting overwritten, before they happen.
Stop prompting and praying. Start forcing your AI to plan.
Frequently asked questions
What does “prompting and praying” mean?
Typing a vague prompt, hitting enter and hoping for the best. With a chatbot you can just regenerate. With an autonomous agent that can read your files, control your browser and run unattended for hours, a vague prompt and a walk-away can end in what Sandeep calls a very expensive mess.
What is plan mode for an AI agent?
Plan mode forces the AI to outline what it intends to do before it acts: the steps, the files it will modify, the tools it will use and where its verification checks are. You review and approve that plan first.
Why approve an AI agent’s plan before it starts?
A plan exposes the agent’s assumptions before it touches anything. Reviewing it lets you place the guardrails and approve the route, which helps you catch runaway processes, a burned API budget and overwritten files before they happen.
Related Links
- Watch the episode on YouTube
- Previous episode: Why the “Harness” Matters More Than the AI Model
- The Danger of “YOLO Mode”
- Stop Writing Prompts (Do This Instead)
Brains Bots and Business is where Sandeep shows you how to actually use AI to grow your company.
Sandbox Media is a marketing and AI agency in Milton, Ontario, serving businesses across the GTA and Canada.